The Goa digital-arrest case began with a woman being deceived through a screen. But as the investigation progressed, the trail led to bank accounts, cash, companies, money changers and financial transactions. The screen was only the beginning.
Nilesh Mathur
I can now manufacture convincing voices, faces, and videos.
But the bigger danger may be more unsettling: when anything can be faked, even genuine evidence can be dismissed as fake.
For decades, the most basic test of reality was also the simplest: seeing was believing.
A photograph showed what happened. A video captured an event. A familiar voice on the telephone established who was speaking. A person appearing on a video call was, by and large, the person on the other side.
That certainty is disappearing.
In Goa, a woman was allegedly subjected to what has come to be known as a “digital arrest”. She was made to believe that she was under investigation, kept under continuous surveillance over a video call, and coerced into transferring ₹2.60 crore between May 21 and June 2, 2025, to accounts falsely described as “Secret Supervision Accounts”, according to the Enforcement Directorate (ED).
What began as a cyber-fraud case has since opened a window into something much larger.
The ED arrested Fahim Moin Hussain Sayed and Naim Mueen Sayyed on August 23, 2026, under the Prevention of Money Laundering Act. The agency says the victim’s money moved within hours through dormant and newly opened bank accounts before being fragmented across more than 400 beneficiary accounts.
The investigation has since widened. On August 27, the ED arrested Bhushan Suryakant Moye, Vilas Narayan Pawar and Shailesh Dagdu Chavan, taking the total number of arrests in the case to five.
According to the ED, the money entered an organised apparatus that received proceeds of cyber fraud as banking credits, converted them into cash and then into foreign currency through companies holding full-fledged money changer licences of the Reserve Bank of India.
The agency says the funds were fragmented across more than 400 beneficiary accounts and passed into companies whose directors on record were people of no means, while others operated the accounts. Accounts linked to these entities are the subject of 330 victim complaints and 163 FIRs across more than 20 states and Union Territories, involving reported losses of ₹417.49 crore. The entities had carried out banking transactions exceeding ₹27,850 crore.
The ED says Moye, a chartered accountant, incorporated 21 companies using identity documents supplied to him and handled those companies and their 43 dummy directors as one group. Pawar allegedly arranged the RTGS entries through which tainted funds reached the network’s accounts, while Chavan allegedly supplied cash against those transfers. Police searched the premises of all three on August 21.
The Goa case is not, in itself, a case about an AI-generated video.
That is precisely why it matters.
The fraud did not require a spectacular, flawless deepfake. It required a convincing digital interaction, the authority of a screen, and enough psychological pressure to make the victim act.
Artificial intelligence is now adding a powerful new layer to that old formula.
The problem is no longer simply that a machine can make something fake look real.
It is that things we once trusted because they looked real may no longer be enough to establish what actually happened.
THE BARRIER HAS DISAPPEARED
A few years ago, creating convincing synthetic media was largely a specialist exercise. It required technical knowledge, computing resources and time.
That barrier has largely disappeared.
Prof. Dinesh K. Vishwakarma, Head of the Department of Information Technology at Delhi Technological University (DTU), says generative AI has automated much of the technical complexity involved in creating synthetic media. Tools can now generate or modify images, clone voices, alter faces and synchronise lip movements with relatively little expertise.
Older deepfakes often revealed themselves through unnatural expressions, poor lip-sync, strange eye movements or distorted backgrounds. But as generative models improve, those familiar giveaways are becoming less reliable.
The problem is made more complicated by partially manipulated material, where most of a video is genuine, but a face, voice, or particular statement has been altered.
Mayank Talwar, Co-founder, EnfoldAI, says voice is among the easiest things to clone. Thirty to 60 seconds of someone’s audio — from an Instagram reel, wedding video or webinar — can be enough to create a convincing voice clone. A fraudster may not need a perfect, full-length video either. A shaky ten-second clip on a poor connection can be enough, because poor quality itself can explain away glitches.
Talwar says his company tested the process with someone without a technical background who created a usable voice clone in less than 15 minutes on the first attempt.
The technological shift is not merely better deepfakes. It is the removal of expertise as a gatekeeper.
And AI is rarely operating alone.
Talwar says what is emerging is a social-engineering operation in which AI is one component. Fraudsters may first study a target through LinkedIn, company websites, Instagram, or public filings, then build a lookalike website, a WhatsApp Business profile carrying a genuine logo, or a spoofed caller ID.
AI enters when doubt needs to be removed: a short voice note in the boss’s voice or a brief video call in which a senior executive or uniformed official appears on screen.
Then comes the old-fashioned part — urgency, secrecy and pressure. The payment goes to a mule account and can be moved within minutes.
The deepfake, in other words, is not necessarily the fraud.
It is the piece that closes it.
WHY WE ARE SO READY TO BELIEVE
Technology explains how a fake can be created. Psychology explains why it can travel.
Shilpi Saraswat, clinical psychologist at Sakra World Hospital in Bengaluru, says people often find it cognitively easier to accept information than to reject it. Repetition is one reason. The illusory truth effect means repeated exposure creates familiarity, and the brain can mistake that familiarity for accuracy.
Emotion makes the problem worse. Fear, anger, and outrage can produce an immediate reaction and reduce the likelihood that someone will stop and examine what they have just seen.
Then there is trust. A WhatsApp message from a friend or relative does not arrive as entirely anonymous information. The relationship itself can act as an implicit endorsement.
Saraswat also points to a deeper mechanism: misinformation often works because it fits existing beliefs and identities. Once a fabricated claim fits comfortably within an existing worldview, correcting it can become surprisingly difficult.
Dr. Kalpana Srivastava, Director, Amity Institute of Behavioural Health and Allied Sciences, says video itself can carry psychological authority because it gives people the feeling that they are directly witnessing something. A face, voice, expressions, and movement create a powerful sense of authenticity.
But seeing something does not establish when, where, why, or under what circumstances it happened.
A genuine video can be edited, taken from another event, or presented with a misleading caption. Increasingly realistic manipulated and AI-generated content adds another layer of uncertainty.
Nor does intelligence provide complete protection. Education can help people evaluate information, Srivastava says, but it does not make anyone immune to cognitive biases. People interpret new information through existing beliefs, experiences and expectations. Intelligence can sometimes even make a person better at constructing arguments for a conclusion they already prefer.
The question, therefore, is not simply whether someone is intelligent enough to spot misinformation. It is whether they pause, question the claim, and examine the evidence.
WHEN THE VIDEO IS REAL BUT THE CLAIM IS FAKE
Not everything misleading is artificially generated.
Pratyush Ranjan, Chief Editor – Digital Services, AI Integration & Fact Check, Press Trust of India (PTI), says traditional misinformation remains widespread even as AI-generated images, videos and synthetic audio become more common.
A PTI Fact Check case involving Union minister Piyush Goyal demonstrates the problem. A video circulated carrying the PTI logo and a claim that Goyal had allegedly threatened protesters and called for some of them to be hanged.
The fact-check team traced a visually identical video to PTI’s official X account, where it had been published on July 23. The visuals matched, but the audio did not. The original footage showed Goyal discussing the government’s response to examination paper leaks, while the viral version carried different audio.
The team analysed the audio with an AI detection tool, which indicated that it was AI-generated. Goyal later said the video had been maliciously manipulated using AI and that an FIR had been registered.
The face was real. The surroundings were real. The recording was real.
But the audio had been manipulated.
The question for fact-checkers is therefore no longer simply whether a claim is true. It can first be phrased as: Is the evidence presented to support the claim genuine?
HOW INVESTIGATORS ESTABLISH WHAT IS REAL
For the police, establishing whether digital content is genuine is now much more complicated than looking for a distorted face.
Gaurav Tyagi, DCP, IFSO, Special Cell, Delhi Police, says investigators use a combination of technical examination and traditional investigative methods.
They may examine metadata, timestamps, device information, software signatures, compression artefacts and encoding inconsistencies. Investigators can examine video frame by frame for abnormalities in facial movements, lighting, shadows, lip-sync, or manipulated objects. Audio can be subjected to spectral analysis, including examination of inconsistencies in ambient sound.
But no single clue is conclusive.
Investigators also try to trace where the content first appeared and identify the earliest available version. Reverse image and video searches can help. If the content relates to a real event, investigators can check it against CCTV footage, call records, eyewitness accounts, and other independent evidence.
Where necessary, material can be sent to forensic laboratories.
The objective is not to find one magical sign of fakery. It is to build a chain of evidence.
And that chain becomes weaker every time the file is altered.
THE COPY-OF-A-COPY PROBLEM
A video downloaded from social media is not necessarily the same evidence as the original.
Compression, cropping, resizing, re-encoding, and repeated forwarding can strip metadata and introduce new digital artefacts. A screen recording is particularly problematic because it creates a new recording of what was displayed rather than preserving the original file.
Tyagi says a screen-recorded video can lose the original file signature entirely. Repeated compression can also destroy forensic details investigators might otherwise use.
The paradox is striking.
The more a piece of misinformation spreads, the more its original forensic trail may disappear.
By the time it reaches millions of people, investigators may be looking at a degraded copy of a copy.
WHEN THE FAKE MAKES THE REAL LOOK FAKE
The AI era has an even more unsettling consequence.
Suppose a genuine video exposes wrongdoing. In the past, simply denying the video might have seemed implausible.
Now, someone can say: “It is AI.”
Tyagi says investigators have encountered cases in which genuine audio or video evidence has been challenged as AI-generated, even when other evidence supports its authenticity.
Ranjan describes this as the “liar’s dividend”—once society knows convincing synthetic content exists, people can use that knowledge to dismiss authentic evidence.
That creates two ways to lose the truth.
People can believe something that never happened.
Or they can refuse to believe something that did.
AUTHENTICATION, NOT JUST DETECTION
Detection technology is improving. But keeping pace with generative AI is difficult.
Prof. V. Ramgopal Rao, Group Vice-Chancellor, Birla Institute of Technology & Science (BITS) Pilani, argues that verification cannot be left entirely to ordinary citizens. People should not have to become forensic experts before deciding whether a video or audio recording is genuine.
Authentication needs to be built into the technology itself.
Cameras and recording devices could cryptographically sign content when it is captured. Secure timestamps, tamper-evident metadata, and a verifiable record of subsequent editing could create a chain of provenance.
Watermarks can help, but they cannot be the only safeguard because they can be removed or manipulated.
The shift is fundamental: from asking whether something looks real to asking whether its origin and integrity can be established.
THE DANGER OF TRUTH FATIGUE
There may be an even quieter consequence of living in an environment saturated with synthetic and misleading content.
Srivastava says repeated exposure to conflicting claims, misleading posts, fact-checks, and counterclaims can become exhausting. Some people may eventually stop asking which version of a story is true and conclude that nobody really knows — or that everything is probably manipulated. That creates a danger beyond believing a particular falsehood.
Misinformation can make people tired of searching for the truth itself.
A person who believes everything can potentially be corrected with evidence. Someone who believes nothing may reject credible journalism, scientific evidence, and expert knowledge along with fabricated content.
WHAT SHOULD AN ORDINARY PERSON DO?
The first step is the simplest: pause.
Don’t forward shocking material immediately. Check whether credible news organisations or official sources are reporting the same event. Look at the original account or number. Reverse-search an image or video where possible. Check whether an older or longer version exists. Pay particular attention to messages designed to provoke fear, anger, shock or urgency.
If someone you know suddenly calls or sends a voice message asking for money or some unusual action, do not verify them within the same conversation. Call them back on a number you already know.
And don’t assume that because AI can create a fake, every disputed piece of evidence is fake.
That is precisely the trap the technology creates.
THE NEW MEANING OF EVIDENCE
The Goa digital-arrest investigation began with a woman being deceived on a screen. It then moved into the physical world — through bank accounts, cash withdrawals, companies, money changers and financial trails.
The screen was only the beginning. The same is true of misinformation.
The damage caused by a fake video does not remain confined to the device on which it is watched. It can damage reputations, facilitate fraud, distort public debate, and alter how people understand a real event.
The technology will continue to improve. So will the methods used to manipulate it.
The durable answer is not simply to teach everyone how to spot a deepfake. It is to build systems in which authenticity can be demonstrated rather than guessed.
For journalists, that means establishing sources and provenance. For investigators, preserving original files and corroborating digital evidence. For technology companies, building authentication into devices and platforms. For governments, developing rules and enforcement mechanisms that can keep pace with the technology.
And for ordinary users, it means learning one new habit:
Don’t confuse familiarity with authenticity.
A face can be real and the voice fake.
A video can be genuine and the context false.
A recording can be authentic and the accompanying claim fabricated.
And genuine evidence can now be dismissed simply because everyone knows that convincing fakes exist.
The old world gave us a simple equation: seeing was believing.
The digital world no longer allows that luxury.
The replacement is less convenient and far more demanding:
Verify the source. Check the context. Establish the provenance. Corroborate the evidence.
In the age of generative AI, trust can no longer rest on what the eye sees or the ear hears.
It has to rest on what can be verified.
